[ODE] redundant constrain problems?

[Martin C. Martin]

Hey Graham,

I'm a little fuzzy about your problem, but if I understand it correctly, 
it's not a problem with ODE but a problem with the way you're setting up 
the problem.  It sounds like you have inconsistent constraints.

For example, if you connect two bodies using two sliders at exactly the 
same attachment points, and you set one slider to 5 cm and the other to 
10, you'll get really high forces as they "fight" over the correct length.

Or is it more like setting them both to 5 cm and getting weird results?

On possibility is to make your joints a little "sloppy," i.e. tolerate a 
little error.  That might fix things up.

- Martin

Graham Fyffe wrote:
> Hi gang.  I have noticed various problems arising from having
> redundant constraints in my joint setups.  My limited understanding of
> WorldStep tells me that this has to do with having the number of
> unconstrained variables pretty much hard coded into the joint classes.
>  This basically pushes the burden of figuring out how many
> unconstrained variables are in the whole system onto the user, who has
> to be very careful about making accidentally redundant constraints.
> 
> The results that I'm seeing with redundant constraints on my car
> suspension is that the car will lift into the air and the whole
> chassis will begin to roll left-over-right, as if some large external
> torque were being applied to it.  This problem goes away if I tweak my
> joint configuration a little bit, but this is unstable at best when
> the user has the freedom to make suspension adjustments :P
> 
> Especially aggravating would be a joint system without redundant
> constraints that is manipulated in such a way as to have redundant
> constraints.  I'm not sure if this can ever really happen.
> 
> So, what I'm getting at, is there any way to modify WorldStep so that
> it doesn't rely on the info from the joints about how many
> unconstrained variables there are, and instead directly calculate it
> from the constraint matrix?  Or is this just way too expensive?
> 
> Thanks everybody!
> 
> - Graham
> 
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode